// MCP TOKEN COST

The Hyperion MCP server costs 15,496 tokens before the first call.

Every request your agent makes carries every tool definition this server exposes — context your code, documents and conversation can't use, mostly for tools the agent never calls. You don't need them all in the window, and you don't have to pay for them.

QUICK ANSWER The Hyperion MCP server's 355 tool definitions consume 15,496 tokens — 7.7% of a 200k context window, and 8.3× the median MCP server (1,862 tokens). A scoped grant exposing only the tools you use cuts that roughly in proportion.

MEASURED FROM SCHEMAS tiktoken o200k_base · rank #1176 of 12,722 measured servers · refreshed every build Method →

// CONTEXT WINDOW SHARE

What that costs before your agent starts working.

Tool definitions are overhead: they occupy context on every request and compete with your code, documents and conversation history for the same window.

200K WINDOW 7.7%

1M WINDOW 1.5%

Corpus context: Hyperion ranks #1176 of 12,722 measured MCP servers by definition cost. The median is 1,862 tokens, p90 is 13,674, and the heaviest (Ainumbers Mcp Apps) is 442,047 — 221% of a 200k window on its own. New to this? See MCP token cost and context window in the glossary.

Token cost is one axis. See the risk picture across your whole stack →

// PER-TOOL BREAKDOWN

Where the 15,496 tokens go.

Each row is one tool definition as a tools/list entry — name, description and input schema — counted with o200k_base. Average: 44 tokens per tool.

Tool	Category	Tokens	% of server
math	Read	167	1.1%
chart	Read	159	1.0%
http_client	Read	145	0.9%
ascii	Read	127	0.8%
data_convert	Read	118	0.8%
hash	Read	117	0.8%
regex	Read	115	0.7%
netutils	Read	109	0.7%
mcp_discover	Read	108	0.7%
encode	Read	104	0.7%
qr	Read	103	0.7%
cron	Execute	91	0.6%
summarize	Read	89	0.6%
web_scrape	Read	88	0.6%
search	Read	84	0.5%
diff	Read	75	0.5%
mcp_readme	Read	65	0.4%
execute_js	Execute	61	0.4%
re_report	Read	60	0.4%
o_intelx_email	Read	53	0.3%
o_airquality	Read	52	0.3%
o_leaklookup	Read	51	0.3%
o_snusbase_hash	Read	51	0.3%
lynx_inspect	Read	50	0.3%
lynx_recon	Read	50	0.3%
o_hibp	Read	50	0.3%
o_jarm	Read	49	0.3%
o_paste_email	Read	49	0.3%
o_robotsmeta	Read	49	0.3%
o_cryptomarket	Read	48	0.3%
o_doh	Read	48	0.3%
o_emailrep	Read	48	0.3%
o_hibp_email	Read	48	0.3%
o_punycode	Read	48	0.3%
o_breach_aggregator	Read	47	0.3%
o_ipv4classify	Read	47	0.3%
o_smtpbanner	Read	47	0.3%
o_trufflehog_url	Read	47	0.3%
o_casify	Read	46	0.3%
o_cve_timeline	Read	46	0.3%
o_disasters	Read	46	0.3%
o_hashtext	Read	46	0.3%
o_internetdb	Read	46	0.3%
o_macvendor	Read	46	0.3%
o_numlookup	Read	46	0.3%
o_paste_domain	Read	46	0.3%
o_quakes	Read	46	0.3%
o_urlscan	Read	46	0.3%
o_whoisserver	Read	46	0.3%
re_search	Read	46	0.3%
lynx_steg	Read	45	0.3%
o_api_key_scan	Read	45	0.3%
o_atbash	Read	45	0.3%
o_cve_poc_checker	Read	45	0.3%
o_dns	Read	45	0.3%
o_exploit_cve	Read	45	0.3%
o_latlonformat	Read	45	0.3%
o_npmdownloads	Read	45	0.3%
o_phonefmt	Read	45	0.3%
titan_generate	Write	45	0.3%
o_base36	Read	44	0.3%
o_caesar	Read	44	0.3%
o_cidr	Read	44	0.3%
o_cve	Read	44	0.3%
o_cvedetail	Read	44	0.3%
o_cve_severity	Read	44	0.3%
o_gh_dorking	Read	44	0.3%
o_goproxy	Read	44	0.3%
o_greynoise	Read	44	0.3%
o_hexdump	Read	44	0.3%
o_homoglyph	Read	44	0.3%
o_hudsonrock	Read	44	0.3%
o_imagerev	Read	44	0.3%
o_ip_math	Read	44	0.3%
o_leakcheck	Read	44	0.3%
o_onthisday	Read	44	0.3%
o_rdap	Read	44	0.3%
o_revimg	Read	44	0.3%
o_suntimes	Read	44	0.3%
o_teamcymru	Read	44	0.3%
o_utm	Read	44	0.3%
o_wayback_leaks	Read	44	0.3%
o_wordcount	Read	44	0.3%
re_summarize	Read	44	0.3%
o_dockerhub	Write	44	0.3%
mcp_stats	Read	43	0.3%
o_btcfees	Read	43	0.3%
o_clickjacking	Read	43	0.3%
o_dnsbl	Read	43	0.3%
o_ean	Read	43	0.3%
o_feodoips	Read	43	0.3%
o_flightsnear	Read	43	0.3%
o_gh_secret_scan	Read	43	0.3%
o_gleif_name	Read	43	0.3%
o_mgrs	Read	43	0.3%
o_morse	Read	43	0.3%
o_osv	Read	43	0.3%
o_phonepivot	Read	43	0.3%
o_railfence	Read	43	0.3%
o_regdomain	Read	43	0.3%
o_roman	Read	43	0.3%
o_s3buckets	Read	43	0.3%
o_unicode_lookup	Read	43	0.3%
o_wikipedia	Read	43	0.3%
v_hash	Read	43	0.3%
o_asnlookup	Read	42	0.3%
o_barcode	Read	42	0.3%
o_circlhash	Read	42	0.3%
o_color	Read	42	0.3%
o_cratestats	Read	42	0.3%
o_dnsverify	Read	42	0.3%
o_github_code	Read	42	0.3%
o_hstspreload	Read	42	0.3%
o_jslibs	Read	42	0.3%
o_marineweather	Read	42	0.3%
o_nvdcve	Read	42	0.3%
o_pluscode	Read	42	0.3%
o_revgeo	Read	42	0.3%
o_srvlookup	Read	42	0.3%
o_subbrute	Read	42	0.3%
o_tldinfo	Read	42	0.3%
v_case	Read	42	0.3%
o_base_convert	Write	42	0.3%
o_formaudit	Write	42	0.3%
fc_scrape	Read	41	0.3%
o_abusecontact	Read	41	0.3%
o_agify	Read	41	0.3%
o_antipode	Read	41	0.3%
o_arxiv	Read	41	0.3%
o_breachdirectory	Read	41	0.3%
o_checksum	Read	41	0.3%
o_decode	Read	41	0.3%
o_dehashed_domain	Read	41	0.3%
o_dnsprop	Read	41	0.3%
o_genderize	Read	41	0.3%
o_geocode	Read	41	0.3%
o_ghkeysgpg	Read	41	0.3%
o_huggingface	Read	41	0.3%
o_ipint	Read	41	0.3%
o_kev	Read	41	0.3%
o_macvendorlookup	Read	41	0.3%
o_nationalize	Read	41	0.3%
o_nstrace	Read	41	0.3%
o_openalexwork	Read	41	0.3%
o_password_dumps	Read	41	0.3%
o_phoneapps	Read	41	0.3%
o_phonevcard	Read	41	0.3%
o_rot47	Read	41	0.3%
o_spacepeople	Read	41	0.3%
o_swagger	Read	41	0.3%
o_uuid	Read	41	0.3%
o_wikidata	Read	41	0.3%
v_json_diff	Read	41	0.3%
v_slug	Read	41	0.3%
o_csp_parse	Execute	40	0.3%
o_transform	Execute	40	0.3%
echo_tts	Read	40	0.3%
o_binarytext	Read	40	0.3%
o_cloud	Read	40	0.3%
o_crossrefauthor	Read	40	0.3%
o_elevation	Read	40	0.3%
o_hnuser	Read	40	0.3%
o_holidays	Read	40	0.3%
o_htmlcomments	Read	40	0.3%
o_ipfull	Read	40	0.3%
o_ipgeo	Read	40	0.3%
o_isin	Read	40	0.3%
o_leetcode_user	Read	40	0.3%
o_maidenhead	Read	40	0.3%
o_maven	Read	40	0.3%
o_metatags	Read	40	0.3%
o_musicbrainz	Read	40	0.3%
o_nato	Read	40	0.3%
o_openfoodfacts	Read	40	0.3%
o_otxdomain	Read	40	0.3%
o_otxip	Read	40	0.3%
o_passgen	Read	40	0.3%
o_permissions_pol	Read	40	0.3%
o_phonecc	Read	40	0.3%
o_phonespam	Read	40	0.3%
o_referrer_pol	Read	40	0.3%
o_sri_check	Read	40	0.3%
o_supply_chain_risk	Read	40	0.3%
o_tls	Read	40	0.3%
o_tlsscan	Read	40	0.3%
o_unpaywall	Read	40	0.3%
o_urlscansearch	Read	40	0.3%
o_useragent	Read	40	0.3%
o_weather	Read	40	0.3%
o_whois	Read	40	0.3%
o_wikidatasearch	Read	40	0.3%
titan_order	Read	40	0.3%
mobsf_report	Read	39	0.3%
o_attack_surface	Read	39	0.3%
o_base32	Read	39	0.3%
o_base64	Read	39	0.3%
o_bgphistory	Read	39	0.3%
o_btcaddr	Read	39	0.3%
o_cfradar	Read	39	0.3%
o_cookies	Read	39	0.3%
o_dnsmx	Read	39	0.3%
o_emailsec	Read	39	0.3%
o_favicon	Read	39	0.3%
o_gravatarfull	Read	39	0.3%
o_httpcode	Read	39	0.3%
o_httpmethods	Read	39	0.3%
o_ioc_reputation	Read	39	0.3%
o_isotime	Read	39	0.3%
o_links	Read	39	0.3%
o_linktree	Read	39	0.3%
o_macvalid	Read	39	0.3%
o_moonphase	Read	39	0.3%
o_npmpkg	Read	39	0.3%
o_openalexauthor	Read	39	0.3%
o_pgp	Read	39	0.3%
o_phishing_intel	Read	39	0.3%
o_portlookup	Read	39	0.3%
o_pubmed	Read	39	0.3%
o_rdapip	Read	39	0.3%
o_revgeocode	Read	39	0.3%
o_rubygemrev	Read	39	0.3%
o_semanticscholar	Read	39	0.3%
o_sopostuser	Read	39	0.3%
o_tlsa	Read	39	0.3%
o_uuid_validate	Read	39	0.3%
o_waf	Read	39	0.3%
re_semantic	Read	39	0.3%
styx_renew	Read	39	0.3%
titan_beacon	Read	39	0.3%
v_b64dec	Read	39	0.3%
v_baseconv	Read	39	0.3%
v_urlparse	Read	39	0.3%
echo_order	Read	38	0.2%
mobsf_scans	Read	38	0.2%
o_ascii85	Read	38	0.2%
o_bimi	Read	38	0.2%
o_breachsearch	Read	38	0.2%
o_c2_infrastructure	Read	38	0.2%
o_cratedownloads	Read	38	0.2%
o_credential_stuffing	Read	38	0.2%
o_datacite	Read	38	0.2%
o_depsdev	Read	38	0.2%
o_dirlisting	Read	38	0.2%
o_dnsquery	Read	38	0.2%
o_dnsrecon	Read	38	0.2%
o_ens	Read	38	0.2%
o_epss	Read	38	0.2%
o_hackernews	Read	38	0.2%
o_ipwhois	Read	38	0.2%
o_isbn	Read	38	0.2%
o_isexitnode	Read	38	0.2%
o_manifest	Read	38	0.2%
o_mimetype	Read	38	0.2%
o_musicbrainzartist	Read	38	0.2%
o_pypiproject	Read	38	0.2%
o_quoted_printable	Read	38	0.2%
o_rpki	Read	38	0.2%
o_securitytxt	Read	38	0.2%
o_stackoverflow	Read	38	0.2%
o_telegram_channel	Read	38	0.2%
o_threatcrowd	Read	38	0.2%
o_tor	Read	38	0.2%
o_wellknown	Read	38	0.2%
o_wikisummary	Read	38	0.2%
o_zenodo	Read	38	0.2%
px_order	Read	38	0.2%
styx_beacon	Read	38	0.2%
v_cron	Read	38	0.2%
v_json_csv	Read	38	0.2%
v_metrics	Read	38	0.2%
o_wikipv	Write	38	0.2%
o_asn	Read	37	0.2%
o_asrank	Read	37	0.2%
o_cdnjs	Read	37	0.2%
o_codeberg	Read	37	0.2%
o_codeforces	Read	37	0.2%
o_cors	Read	37	0.2%
o_doi	Read	37	0.2%
o_feeds	Read	37	0.2%
o_githubsearch	Read	37	0.2%
o_golangpkg	Read	37	0.2%
o_hashid	Read	37	0.2%
o_infra_fingerprint	Read	37	0.2%
o_leet	Read	37	0.2%
o_nearbywiki	Read	37	0.2%
o_npmorg	Read	37	0.2%
o_nuget	Read	37	0.2%
o_numwords	Read	37	0.2%
o_openphish	Read	37	0.2%
o_peers	Read	37	0.2%
o_port	Read	37	0.2%
o_pypistats	Read	37	0.2%
o_rdap_domain	Read	37	0.2%
o_robots	Read	37	0.2%
o_rubygems	Read	37	0.2%
o_slug	Read	37	0.2%
o_steam_user	Read	37	0.2%
o_urlparse	Read	37	0.2%
px_info	Read	37	0.2%
px_plans	Read	37	0.2%
styx_fetch	Read	37	0.2%
v_epoch	Read	37	0.2%
v_regex	Read	37	0.2%
v_urldec	Read	37	0.2%
o_datauri	Write	37	0.2%
echo_beacon	Read	36	0.2%
fc_map	Read	36	0.2%
o_blockheight	Read	36	0.2%
o_breachdb	Read	36	0.2%
o_chesscom	Read	36	0.2%
o_crc32	Read	36	0.2%
o_crypto	Read	36	0.2%
o_devto	Read	36	0.2%
o_dnsgraph	Read	36	0.2%
o_email	Read	36	0.2%
o_hnsearch	Read	36	0.2%
o_keybase	Read	36	0.2%
o_lobsters	Read	36	0.2%
o_npmdl	Read	36	0.2%
o_pypi	Read	36	0.2%
o_rot13	Read	36	0.2%
o_semver	Read	36	0.2%
o_sitemap	Read	36	0.2%
o_social	Read	36	0.2%
o_stringmetrics	Read	36	0.2%
o_trendingwiki	Read	36	0.2%
o_uuencode	Read	36	0.2%
v_b64enc	Read	36	0.2%
v_hex2txt	Read	36	0.2%
v_jwt	Read	36	0.2%
fc_search	Read	35	0.2%
o_bluesky	Read	35	0.2%
o_crates	Read	35	0.2%
o_epoch	Read	35	0.2%
o_headers	Read	35	0.2%
o_hostname	Read	35	0.2%
o_htmlencode	Read	35	0.2%
o_httping	Read	35	0.2%
o_medium	Read	35	0.2%
o_mtasts	Read	35	0.2%
o_npm	Read	35	0.2%
o_pageinfo	Read	35	0.2%
o_reversedns	Read	35	0.2%
o_username	Read	35	0.2%
v_htmldec	Read	35	0.2%
v_json_format	Read	35	0.2%
v_now	Read	35	0.2%
v_txt2hex	Read	35	0.2%
v_urlenc	Read	35	0.2%
v_uuid	Read	35	0.2%
o_archiveorg	Read	34	0.2%
o_mac	Read	34	0.2%
o_redirects	Read	34	0.2%
o_tech	Read	34	0.2%
o_gitlab	Read	33	0.2%

// CUT THE COST

Your agent uses a handful of these tools. It pays for all 355.

You don't need all 355 of those definitions in the window. PolicyLayer is an MCP gateway that sits in front of Hyperion: only the tools you grant are exposed to the agent, the rest never load. A smaller window means a sharper agent — less noise when it picks a tool — and every request costs less:

Grant scope	Definition cost	Reduction
All 355 tools (no gateway)	15,496 tokens	—
3 granted tools	~131 tokens	−99%
5 granted tools	~218 tokens	−99%
10 granted tools	~437 tokens	−97%

Create a free account and register Hyperion — nothing to install.
Grant only the tools you use — ungranted definitions never enter the context window.
Point your MCP client (Claude, Cursor, anything) at your gateway URL.

CUT HYPERION TOKEN COST →

Instant setup, no code required.

Model your own stack in the token-cost calculator, or see the Hyperion policy for what a sensible grant looks like.

//FAQ

Hyperion token-cost questions.

How many tokens does the Hyperion MCP server use?+

Its 355 tool definitions total 15,496 tokens — 7.7% of a 200k context window — measured with tiktoken o200k_base over the serialised tools/list payload. Exact counts vary slightly by client and model.

Why does Hyperion consume tokens before I send a message?+

MCP clients load every connected server's tool definitions — name, description, and input schema — into the model's context so it knows what it can call. That payload is charged against your context window on every request, whether or not a tool is used.

How do I reduce Hyperion's token usage?+

Expose fewer tools. A PolicyLayer grant scopes Hyperion to only the tools you allow — ungranted definitions are filtered out of the tool list, so they never enter the context window. A grant of 3 typical tools costs roughly 131 tokens, a 99% reduction.

Does deferred tool loading fix this?+

Partially, in some clients. Claude Code defers MCP tool schemas behind a tool-search step by default, and VS Code has experimental grouping — but you still pay tokens per search and reload, and Cursor, Windsurf and Gemini CLI load definitions upfront. Reducing the exposed tool set cuts the cost in every client.

// METHOD & REVIEW

How these numbers were measured.

01

Serialisation

Each tool is serialised as a tools/list entry — name, description, input schema — from the schemas in the PolicyLayer scan database. Clients differ slightly in framing, so treat counts as close estimates.

02

Tokeniser

tiktoken o200k_base (GPT-4o/o-series). Anthropic's current tokeniser isn't published, so Claude's exact counts will differ; for English text and JSON schemas the totals are close enough to treat these as estimates.

03

Deferred loading

Some clients now defer schema loading (Claude Code's tool search; VS Code experimental grouping). You still pay per search and reload — and Cursor, Windsurf and Gemini CLI load everything upfront.

Computed 20-10-2026 from the PolicyLayer scan database over all 355 catalogued Hyperion tools. Counts refresh with every site build.